Electroencephalograms (EEGs) as a tool to personalized medicine in depression treatments
Conference
65th ISI World Statistics Congress 2025
Format: IPS Abstract - WSC 2025
Keywords: "bayesian, "latent, mixed-effects models
Abstract
The motivation for this study comes from a randomized placebo controlled depression clinical trial of sertraline. Although several factors have been studied in relation to treatment response in depression, it is well known that individual responses to treatment can vary widely. Our dataset includes patients with major depressive disorder who were randomized to receive either the drug or a placebo, and were monitored over an eight-week period. For each subject, several scalar and categorical covariates are available, as well as their resting state electroencephalography (EEG) under a closed eyes condition. The objective is to determine whether EEG measurements can be used to personalize treatment for patients. This EEG data contains the current source density amplitude spectrum values at a total of 14 electrodes located in occipital and parietal brain regions. To effectively model this data, it is crucial to account for the correlation among EEG measurements. This can be done using dimension reduction techniques such as principal component analysis. Although this method is powerful, it does not take advantage of the intrinsic ordering of observations given by the functional structure of the covariate. We propose a functional data approach to analyze the EEG data based on a mixed-effect model across individuals.